INNOVATION OF COLLEGE POP MUSIC
TEACHING IN TRADITIONAL MUSIC
CULTURE BASED ON ROBOT COGNITIVE-
EMOTIONAL INTERACTION MODEL
Hua Zhou*
School of Electronic Engineering, Xi’an Shiyou University, Xi’an, Shaanxi, 710065,
China
Shanghai Salvage, Shanghai, 201914, China
211122@126.com
Reception: 13/11/2022 Acceptance: 06/01/2023 Publication: 21/03/2023
Suggested citation:
Z., Hua. (2023). Innovation of college pop music teaching in traditional
music culture based on robot cognitive-emotional interaction model. 3C
TIC. Cuadernos de desarrollo aplicados a las TIC, 12(1), 200-220. https://
doi.org/10.17993/3ctic.2023.121.200-220
https://doi.org/10.17993/3ctic.2023.121.200-220
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.42 | Iss.12 | N.1 January - March 2023
200
ABSTRACT
Emotional computing and artificial psychology is a new research direction that has
received increasing attention in the field of harmonious human-computer interaction
and artificial intelligence and is also a new intersection of mathematics, information
science, intelligence science, neuroscience, physiology, psychological science and
other multidisciplinary intersection. The current problems and drawbacks in the
teaching of popular music in colleges and universities, and the search for methods
and measures to reform and innovate popular music education in colleges and
universities are the difficulties of current music teaching work. In this paper, we try to
apply a robot cognitive-emotional interaction model to college pop music teaching,
and establish an emotional interaction model based on reinforcement learning with the
help of cognitive-emotional computing of human-computer interaction, to be able to
integrate emotional interaction in pop music teaching and to make an accurate
emotional analysis of students' singing effect. Different from traditional music teaching
methods, the robot-based cognitive-emotional interaction model established in this
paper can establish an innovative teaching model for college pop music teaching and
optimize the teaching effect.
KEYWORDS
Robot cognition; Emotional interaction; Popular music; University teaching;
Optimization and innovation
https://doi.org/10.17993/3ctic.2023.121.200-220
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.42 | Iss.12 | N.1 January - March 2023
201
PAPER INDEX
ABSTRACT
KEYWORDS
1. INTRODUCTION
2. EMOTIONAL DESCRIPTION AND EMOTIONAL INTERACTION
2.1. Affective description model
2.1.1. Dimensional sentiment description model
2.1.2. A discrete sentiment description model
2.2. Cognitive-emotional computing for human-computer interaction
3. THE COGNITIVE-EMOTIONAL INTERACTION MODEL FOR ROBOTS
BASED ON REINFORCEMENT LEARNING
3.1. Cognitive-emotional computing based on reinforcement learning
3.2. Emotional interaction model based on reinforcement learning
3.2.1. Status
3.2.2. Behavior
3.2.3. Discount factor
3.2.4. Rewards
3.2.5. Strategies
3.2.6. Model Optimization
3.2.7. Emotional interaction process simulation
3.3. Robot Emotional State Update
4. INNOVATION OF POPULAR MUSIC TEACHING IN COLLEGES AND
UNIVERSITIES
4.1. Popular music and traditional music culture
4.2. The Role of cognitive-emotional interactive robots in music teaching
5. CONCLUSION
DATA AVAILABILITY
CONFLICTS OF INTEREST
REFERENCES
https://doi.org/10.17993/3ctic.2023.121.200-220
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.42 | Iss.12 | N.1 January - March 2023
202
1. INTRODUCTION
As human communication is natural and contains multiple emotions, it is also
natural to expect computers to have the ability of emotional cognition in human-
computer interaction. How to enable computers to recognize, understand, and
generate human-like emotions has received widespread attention from disciplines
such as computer science, brain science, and psychology, and has given birth to the
intersection of cognitive-emotional computing [1-3]. As an important way to reflect the
application value of artificial intelligence technology, the research of human-computer
interaction systems has received common attention from academia and industry, and
human-computer interaction products have gradually entered people's daily life.
Influenced by the great satisfaction of material needs, people have begun to desire
and pursue the sense of fit brought by human-robot interaction at the emotional level,
and hope that robots can have the cognitive-emotional computing ability to generate
advanced anthropomorphic emotions while satisfying daily interaction needs [4-5].
The study of cognitive-emotional computation for robots is the key to realizing
advanced human-robot interaction technology with organic integration of emotions,
which has important practical significance and high research value. The applications
of robotic cognitive-emotional interaction models are also becoming more and more
multifaceted. In the field of popular music teaching in colleges and universities, this
paper is devoted to making innovations with the help of cognitive-emotional interaction
robots for its disadvantages such as backward teaching model, preaching by the book
and lack of vitality [6].
Intelligent human-robot dialogue systems, as an intuitive manifestation of
intelligence in artificial intelligence technology, have become an important research
component for achieving natural and harmonious human-robot interaction. In recent
years, to improve robot anthropomorphism in human-robot interaction systems and
create a harmonious and friendly human-robot interaction environment, researchers
have conducted a lot of research around cognitive-emotional computing of robots in
open domains, and numerous cognitive-emotional interaction models with important
reference values have emerged. The literature [7-8] proposed an affective interaction
model based on a guided cognitive reappraisal strategy to emotionally regulate
external affective stimuli and promote the positive affective expression of the robot to
some extent. In the literature [9-10], the cognitive emotion model of the robot is
integrated into the smart home environment, and the cognitive reassessment strategy
guided by positive emotion is obtained by optimizing and analyzing the cognitive
emotion model of the service robot in the smart home environment using simulated
annealing algorithm, and the probability of transferring emotional states is updated
based on the cognitive reassessment strategy. Literature [11-12] proposed the multi-
emotion dialogue system MECS, which tends to generate coherent emotional
responses in dialogue and selects the most similar emotion as the robot response
emotion. The literature [13-14] proposes emotional chat machines that can produce
appropriate responses not only in terms of content relevance and syntax but also in
terms of emotional coherence. In response to the teaching of popular music in
colleges and universities, the American Education Act explicitly proposes that music
https://doi.org/10.17993/3ctic.2023.121.200-220
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.42 | Iss.12 | N.1 January - March 2023
203
education in schools should break with tradition and no longer be limited to the
teaching and development of traditional resources, but expand to all aspects of
popular music and contemporary music [15]. German universities believe that both
elegant music and modern music should be actively drawn from the advantageous
factors of these musicians in the teaching and development of universities so that
students can have a richer musical vision from them and can be exposed to richer
musical content [16]. With the continuous development of music education, the
application of popular music in college education has become more and more
extensive, making the music classroom in colleges and universities more colorful and
further highlighting the function and value of popular music teaching.
Establishing and maintaining social connections with others is a basic human need
in interpersonal relationships, and when users use more anthropomorphic robots as
partners and establish relationships with them, they have a better willingness to
continue interacting with these robots for a longer period than with those robots that
are more rigid in their expression. As robots become more anthropomorphic, the user
experience increases and the trust and dependence on the robot increase. In this
paper, we investigate the application of emotional robots to the teaching of popular
music in universities. First, we quantify the emotion analysis of human-robot
interaction by multi-dimensional emotion description, which makes it possible for the
robot to compute emotion. Then, the emotional input and output of human-robot
interaction are simulated based on reinforcement learning algorithms, to establish its
complete cognitive-emotional interaction model. Finally, the model is practiced in
college pop music teaching, and a pop music teaching model that makes innovations
in traditional music culture is proposed.
2. EMOTIONAL DESCRIPTION AND EMOTIONAL
INTERACTION
2.1. AFFECTIVE DESCRIPTION MODEL
The ultimate goal of robotic cognitive emotion computing can be interpreted as
giving human-like artificial emotions to robots in interactive systems by simulating
human emotion processing to build trust between humans and machines. The main
research content includes three parts: emotion recognition, emotion modeling and
emotion understanding. Due to the complexity and abstraction of actual emotion,
before establishing the robot cognitive emotion interaction model, it is necessary to
recognize and quantify the emotion of the user interaction input content, and convert it
into an emotion state vector that can be recognized and processed by the computer,
and at present, according to its different ways of emotion description, the methods for
emotion recognition and quantification The current methods for emotion recognition
and quantification can be divided into dimensional emotion description models and
discrete emotion description models.
https://doi.org/10.17993/3ctic.2023.121.200-220
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.42 | Iss.12 | N.1 January - March 2023
204
2.1.1. DIMENSIONAL SENTIMENT DESCRIPTION MODEL
The dimension-based emotion description model describes emotions as coordinate
points in a state space composed of Cartesian product operations, each dimension in
the state space corresponds to a psychological attribute of emotion, and the
magnitude of the value on the dimension reflects the strength of the emotional
characteristics corresponding to the psychological attribute, and the emotion
description capability of this space covers all emotions [17-18]. In other words, that is,
there is a one-to-one mapping relationship between the emotions that exist in reality
and the coordinate points in the state space. At the same time, the similarities and
differences between different emotions can be quantified and analyzed by calculating
the distance between coordinate points. The basis of emotion calculation is to find this
mapping dimensional theory, which regards the transition between different emotions
as a continuous and smooth state transfer process. The dimensionality-based model
of emotion description has received a lot of attention from scholars because it
combines the characteristics of continuity and complexity of human emotion
distribution. According to the different ways of dividing emotional attributes in
psychology, dimension-based emotion description models have one-dimensional, two-
dimensional, three-dimensional and more multi-dimensional spatial theories.
Currently, the widely used dimensional description models are the two-dimensional
activation-valence spatial theory (AV), where the Arousal axis is the activation
dimension, representing the degree of pleasantness-unpleasantness, and the Valence
axis is the valence dimension, representing the degree of agitation-calmness. The
three-dimensional Pleasure-Activation-Dominance (PAD) spatial theory, in which the
Pleasure axis is the Pleasure dimension, representing the degree of positive and
negative affective states, and the Arousal axis is consistent with the Arousal axis in the
AV spatial theory, representing the individual's level of neurophysiological activation.
The AV spatial theory is shown in Figure 1, from which we can see that the emotional
labels that exist in daily life can be mapped to coordinate points in space, and the
magnitude of the corresponding coordinate values in each dimension varies according
to the strength of each emotional attribute.
Figure 1. Activation Degree-Validity emotional space
pleased
insterested
happy
excited
alamed
JoyAnger
annoyed
nervous
angry
afraied
terrified
furious
Arousal
1
-1
-1
1
Valence
content
relaxed
serenecalm
NeuralSadness
sad
depresed
boredsleepy
https://doi.org/10.17993/3ctic.2023.121.200-220
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.42 | Iss.12 | N.1 January - March 2023
205
2.1.2. A DISCRETE SENTIMENT DESCRIPTION MODEL
Discrete models of emotion description describe emotions in the form of a finite
number of adjective labels and are widely used in people's daily lives, as well as in
early computational research on emotions. Rich linguistic labels can describe a large
number of affective states, but which classifications are of higher research value? This
question can be attributed to the classification of basic affective states based on
adjectival labels [19-21]. In general, the categories of emotions that can cross different
races and cultures and are shared by humans and socially oriented mammals are the
basic emotions. Table 1 lists the definitions and classifications of basic emotions by
different scholars.
Table 1. Definition of basic emotions by different scholars
In summary, considering the rich and strong emotions of participants in open-
domain HCI systems, the continuous model based on the PAD emotion space is
chosen for the quantitative analysis of emotions in the open-domain cognitive emotion
study of robots [22-24]. In closed-domain HCI systems, the emotions of the
participants are simpler, and the discrete emotion model is chosen to classify the
emotions into positive, negative and neutral when studying the cognitive emotions of
robots for the closed domain.
2.2. COGNITIVE-EMOTIONAL COMPUTING FOR HUMAN-
COMPUTER INTERACTION
In the long history of robotics research, researchers have focused more on the
design and manufacturing of robots, control systems, drive systems, and content
representation, until the introduction of "affective computing" and the gradual shift of
Scholars Basic emotions
Weiner, Graham Happiness, Sadness
Watson Fear, Love, Rage
James Fear, Grief, Love, Rage
Panksepp Anger, Disgust, Anxiety, Happiness, Sadness
Ekman, Friesen,
Ellsworth Anger, Disgust, Fear, Joy, Sadness, Surprise
Fridja Desire, Happiness, Interest, Surprise, Wonder, Sorrow
McDougall Fear, Disgust, Elation, Fear, Subjection, Tender-emotion,
Wonder
Plutchik Anger, Interest, Contempt, Disgust, Distress, Fear, Joy,
Shame,
Surprise
Tomkins Anger, Interest, Contempt, Disgust, Distress, Fear, Joy,
Shame,
Surprise
Arnold Anger, Aversion, Courage, Dejection, Desire, Despair,
Dear, Hate,
Hope, Love, Sadness
https://doi.org/10.17993/3ctic.2023.121.200-220
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.42 | Iss.12 | N.1 January - March 2023
206
robotics research to intelligent robotics, robot cognitive emotion research has received
increasing attention [25-27]. The study of robot cognitive emotion has received
increasing attention from researchers [25-27]. Emotion is a subjective response to a
valued relationship, which is simply the process of perceiving the impact of external
values on oneself, and to facilitate an intuitive understanding of this process, a triad
can be used to represent cognitive affective computing formally:
SC = (S, C, W) (1)
where S denotes the set composed of different types of information carriers such
as text, speech and pictures. Different carriers contain different emotional features,
and there are large differences between the representations of different features.
C denotes the set composed of different sentiment categories. C can denote a
discrete set of sentiment states composed of several basic sentiment states, or a
spatial set of sentiment states composed of different sentiment dimensions.
W denotes the set composed of different emotional trait intensities, and the
intensities can be initially quantified into basic high, medium and low levels, or further
subdivided.
Emotional features combined with intensity features form the core of cognitive
sentiment computing. Cognitive sentiment computing first identifies and quantitatively
analyzes the data features related to objective things extracted from different
information carriers. Secondly, the affective features in the data features are
calculated under different polarity dimensions to achieve the subjective affective
perception of objective things. Finally, it is fed back to the participant with an
appropriate sentiment expression. The cognitive sentiment computation can be further
described as a combination of state space composed of S, C and W by Cartesian
product operation, i.e:
SC = S × C × W (2)
In recent years, the emergence of smart speakers and other devices has largely
promoted the productization of human-robot interaction systems. In the process of
cognitive-emotional modeling of robots, existing research has achieved better results
in the extraction and recognition of emotional features. With the development of
intelligence in various industries, research in AI-related fields has been greatly
promoted, and its ultimate development goal is to establish intelligent bodies with the
ability to observe the environment and to think and decide independently.
https://doi.org/10.17993/3ctic.2023.121.200-220
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.42 | Iss.12 | N.1 January - March 2023
207
3. THE COGNITIVE-EMOTIONAL INTERACTION
MODEL FOR ROBOTS BASED ON REINFORCEMENT
LEARNING
3.1. COGNITIVE-EMOTIONAL COMPUTING BASED ON
REINFORCEMENT LEARNING
In the process of human emotion generation, individual emotional state responses
are not only related to external emotional stimuli but also related to their emotional
states and emotional interaction motives. When performing affective state response,
we should not only consider the influence of contextual multi-round interaction context
on the probability of transferring the current affective state but also consider the
influence of the current affective state response on the subsequent interaction
relationship. Therefore, to effectively carry out robot emotional strategy learning, this
paper proposes to use reinforcement learning features to establish the correlation
between contextual multi-round emotional state and current response emotional state,
and perform cognitive emotion computation for the robot, and the computational
framework is shown in Figure 2.
Figure 2. The framework of robotic affective computing
To facilitate the implementation of participant sentiment state tracking, sentiment
quantification and state evaluation are performed on the interactive input content. In
this paper, we quantify the sentiment of the interactive input content and obtain its
corresponding sentiment value in the PAD continuous sentiment
space. Secondly, the interaction sentiment value vector is evaluated in terms of
state, and its sentiment state vector under the action of six basic sentiment
states in the PAD continuous sentiment space is obtained. The emotion state
evaluation function is defined as:
(3)
(4)
Current emotional
state
Emotional state
transfer probability
External emotional
stimulation
The next moment
of emotional state
Responding to
emotional output
Cognitive-emotional
computing based on
reinforcement learning
Ei= (p,a,d)
I(Ei)
I(Ei)=[i1,i2,i3,i4,i5,i6]
ij=1/h
j=1
61/hj
,hj0
i
1= 0,i2= 0,1,,ij= 1,2,,i6= 0
https://doi.org/10.17993/3ctic.2023.121.200-220
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.42 | Iss.12 | N.1 January - March 2023
208
(5)
where
denotes the interactive input sentiment value. j = 1, 2, …,6 denotes the
six basic sentiment states of happy, surprised, disgusted, angry, fearful, and sad,
respectively; denotes the sentiment value corresponding to basic sentiment j.
denotes the covariance matrix of the clustering region of basic sentiment j. denotes
the distance between and
. then denotes the assessed value of the affective
state of under the action of .
3.2. EMOTIONAL INTERACTION MODEL BASED ON
REINFORCEMENT LEARNING
The reinforcement learning model is based on the principle that intelligence, in its
current state, performs a behavior to interact with the environment and enters a new
state, while obtaining the corresponding immediate reward from the environment, and
then evaluates this behavior according to the reward, and the reward value increases
for behaviors favorable to goal achievement and decays for behaviors unfavorable to
goal achievement, and this process continues to cycle until the termination state
[28-31].
3.2.1. STATUS
The state
indicates the emotional state in which the intelligent body is, which is
usually given by the external environment. To reduce the granularity of emotion
division and increase the continuity and delicacy of robot emotion expression, the
PAD continuous emotion space containing 151 emotion states is taken as the emotion
state space of the intelligent body in this paper, and the emotion state vector of each
emotion state in the space under the action of six basic emotion states is taken as the
possible interaction input response emotion states.
3.2.2. BEHAVIOR
Behavior
denotes an action executed by the intelligent body in the interactive
response process when selecting the next round of response emotional state with the
search space of the emotional space size. The activity process of the intelligent body
in the emotion space is the Markov transfer process between the emotional states in
the emotion space.
3.2.3. DISCOUNT FACTOR
The discount factor γ
can be used to calculate the future reward decay of the
cumulative rewards of a state sequence when the environment is stochastic. In this
paper, we consider that the more distant the future moment is from the current
session, the smaller the effect of future rewards on the satisfaction used to measure
the affective state of the next round of sessions. Its value is between 0 and 1. The
hj= (EiEj)Cj(EiEj)T,j= 1,2,,6
Ej
Gj
hj
Ei
Ej
Ej
s
a
https://doi.org/10.17993/3ctic.2023.121.200-220
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.42 | Iss.12 | N.1 January - March 2023
209
greater the importance of future rewards is considered, the greater the value of , and
vice versa the smaller the value of .
3.2.4. REWARDS
The reward r can be used to measure the future satisfaction of the obtained
affective state after the intelligent body performs the corresponding action
. Both
sides of the human-robot interaction have certain emotional motivations during the
interaction. Based on the principle of interpersonal attraction in social psychology, the
emotional motivation of robot interaction is set to achieve a certain degree of
emotional affirmation, emotional guidance and emotional empathy for the participant,
and the emotional reward function is constructed accordingly.
1.
Similarity emotional reward function: Considering the process of interpersonal
interaction, people often hope that the other party can produce similar emotional
responses to themselves, to achieve the emotional affirmation of the participant, the
cosine similarity is calculated to measure the similarity function between the
emotional state vectors as
(6)
2.
Positive affective reward function: Considering the process of interpersonal
interaction, people will achieve some kind of emotional guidance to others by
adjusting their emotional expression state. Therefore, to achieve emotional
guidance for participants, this paper increases the participants' willingness to
interact by setting the robot's emotional positivity guidance. In fact, the higher the
positivity, the better, especially when the participant's emotion is negative, it may be
counterproductive. The synergistic effect of positivity and similarity can effectively
solve the problem of over-guidance. Therefore, in this paper, the positivity of the
response emotional state vector is calculated as
(7)
3.
Empathic affective reward function: Consider the process of interpersonal
interaction in which interpersonal attraction is not only related to the similarity
between individuals but is also influenced by complementary relationships with
each other. Complementary relationships are influenced by the fact that people
sometimes tend to prefer people who can complement them in some way. In
emotional interactions, this can be interpreted as the expectation that the other
person has empathy and resonates with them in terms of emotional expression.
Therefore, this paper measures emotional empathy by calculating the
interrelationships between emotional state vectors
γ
γ
a
( ) ( ) ( )
( ) ( )
1
1 1
1
,k k
k k
k k
I E I E
r S E E
I E I E
+
+
+
= =
( ) ( )
( )
6
2 1 1
1
k k j j
j
r P E P I E l i
+ +
=
= = =
https://doi.org/10.17993/3ctic.2023.121.200-220
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.42 | Iss.12 | N.1 January - March 2023
210
(8)
For behavior
, the final reward it receives is the weighted sum of the above 3
reward measures
(9)
where , , and are the corresponding weight parameters, respectively.
3.2.5. STRATEGIES
The policy P is used to represent the probability distribution corresponding to when
the intelligence chooses the next emotional state in the current state and can be
expressed by the formula
(10)
The initial value is the initial transfer probability between affective states. The model
is generally optimized using a strategic gradient algorithm, so its value is related to the
future reward value available for selecting the next affective state, with a greater
probability of occurrence for actions that receive a large future reward value and
correspondingly a smaller probability of occurrence for actions that receive a small
future reward value.
3.2.6. MODEL OPTIMIZATION
The model update training is achieved by parameterizing the policy through the
policy gradient algorithm, to maximize the future cumulative reward expectation by
optimizing the model parameters. Therefore, the objective function is to maximize the
expected value of future rewards, defined as
(11)
where denotes the reward value obtained by acting
in the state
; then the gradient is updated using the likelihood ratio technique
(12)
Finally, the parameter is updated using the obtained gradient values
(13)
( ) ( ) ( )
( )
( ) ( )
( )
3 1 2
1
2
1
, log |
1
1log |
1
k k k
k
k
k
r M E E P I E a
rank E
P a I E
rank E
+
+
= = +
++
a
( )
( )
1 1 2 2 3 3
|
k
R a I E r r rα α α= + +
α1
α2
α3
( ) ( )
( )
1
( | ) |
RL k k
a s P I E I Eπ
+
=
( )
( )
( )
1:
1
( ) ,
T
T
RL k k
RH a
k
L E R a I Eθ
=
=
Rk(ak,I(Ek))
ak
I(Ek)
( )
( )
( )
( )
RL 2
1
( ) log , ,
T
k k k k
k
L P a R a
θ θ
θ
=
=
I E I E
θ
RL
( )
new old
L
θ
θ θ β θ= +
https://doi.org/10.17993/3ctic.2023.121.200-220
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.42 | Iss.12 | N.1 January - March 2023
211
When the cumulative reward expectation is maximized, the sentiment state
corresponding to the resulting optimal policy is the optimal response sentiment state
for the interaction input.
3.2.7. EMOTIONAL INTERACTION PROCESS SIMULATION
This chapter uses two Agents to simulate the emotional interaction process
between the Agent and the external environment by continuously interacting with each
other. The interaction process between two Agents can be described as follows: first,
Agent 1 is given a random emotional state
, and then Agent 1 converts it into an
emotional state vector
through emotional evaluation and then transmits this
vector to Agent 2 as input, after which Agent 2 converts the corresponding response
emotional value evaluation into an emotional state vector
back to Agent 1,
and this process is repeated until the maximum number of interaction rounds is
simulated. The interaction goal is to be able to select the optimal emotional state with
the maximum future reward under the current interaction input emotional state. The
emotional interaction process between Agents is shown in Figure 3.
Figure 3. Emotional interaction process
The distance of spatial distance is used to map the similarity between different
affective categories in the affective space. The transfer probability between emotional
states differs with different distances and similarities between categories. The closer
the distance, the greater the state transfer probability; conversely, the farther the
distance, the smaller the state transfer probability. Therefore, to facilitate the
calculation and analysis of the emotional states in response to external emotional
stimuli, this chapter uses the top
emotional states in space that are closest to the
Euclidean distance from the external emotional stimulus point as the candidate
emotional states for each round of Agent interaction.
3.3. ROBOT EMOTIONAL STATE UPDATE
In this paper, the optimal response emotional value of the robot in continuous
emotional space is calculated by combining the six basic emotional values with the
probability of emotional state transfer obtained by the robot after being subjected
to external emotional stimuli to achieve the emotional state transfer of the robot in
E1
(E1)1
E2
I(E2)
E1E3E5
E2E4E6
Agent 1 Agent 1 Agent 1
Agent 2 Agent 2 Agent 2
( )
1
I E
( )
3
I E
( )
5
I E
( )
2
I E
( )
4
I E
Ā
n
Pk+1
RH
Ek+1
RH
https://doi.org/10.17993/3ctic.2023.121.200-220
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.42 | Iss.12 | N.1 January - March 2023
212
continuous emotional space [32-33]. First, assuming that the optimal response
emotional state vector obtained from the reinforcement learning model corresponds to
a strategy
, the probability of transferring the response emotional state to the six
basic emotional states based on the user input can be obtained as:
(14)
Second, the
rounds of robot response emotional state transfer probability
are updated by combining the rounds of robot emotional state transfer
probability and the rounds of interactive user input optimal response emotional
state transfer probability with the following equation:
(15)
(16)
where
represents the confidence level of transferring the interaction input
response emotional state to the 6 basic emotional states [34-35]. The resulting
transfer probability
is then normalized to obtain the transfer probability of the
-round interactive robot response emotional state as:
(17)
Finally, the coordinate position of the
-round robot optimal
response emotional value
in the emotional space is calibrated based on the
obtained robot emotional state transfer probability
, which is calculated as
follows:
(18)
(19)
Based on the above conditions, the interaction input content emotion is quantified
and evaluated based on the PAD emotion space, the user and robot emotion
generation process in the human-robot interaction system is modeled using
reinforcement learning, the long-term correlation between the current interaction input
emotion state and the contextual long-term emotion state is established, and the
model parameters are optimized and updated by maximizing the reward value
p
( )
1 1 2 3 4 5 6
[ (1), (2), (3), (4), (5), (6)] , , , , ,
k
k HR p p p p p p
P E E P P P P P P i i i i i i
+
= =
k1
Pk1
R
k1
Pk+1
RH
k
P(Ek+1 |Ek
HR)
1 1
( ) ( ) ( )
k k
R RH j
P j P j c P j
+
= +
( )
{ }
( )
{ } ( )
{ }
1
1 1
( ) min
max min
k
k HR
jk k
k HR k HR
P j P E E
c
P E E P E E
+
+ +
=
cj
Pk+1
R
k+ 1
[ ]
1
16
1
1
1
1 1 1 1 1 1
( )
( )
( )
(1), (2), (3), (4), (5), (6)
k
R
k
k
R
j
k
RH k k k k k k
P j
P j
P j
P P P P P P P
+
+
+
=
+
+ + + + + +
=
=
(pk+1,ak+1,dk+1)
k+ 1
Ek+1
RH
Pk+1
RH
6
1 1
1
6
1 1
1
6
1 1
1
( )
( )
( )
k j k
j
k j k
j
k j k
j
p p P j
a a P j
d d P j
+ +
=
+ +
=
+ +
=
=
=
=
( )
1
1 1 1
, ,
k
RH k k k
E p a d
+
+ + +
=
https://doi.org/10.17993/3ctic.2023.121.200-220
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.42 | Iss.12 | N.1 January - March 2023
213
expectation, and finally, the optimal response emotion state corresponding to the
obtained user input is realized by updating the transfer probability of k + 1 rounds of
robot emotion state, and the optimal response emotion state of the robot in the
continuous emotion space can be obtained.
4. INNOVATION OF POPULAR MUSIC TEACHING IN
COLLEGES AND UNIVERSITIES
4.1. POPULAR MUSIC AND TRADITIONAL MUSIC CULTURE
In the new era, the traditional music teaching mode is increasingly unable to meet
the continuously growing spiritual and cultural needs of students, and pop music
teaching in colleges and universities needs to develop in a diversified direction so that
students can express their real emotions more intuitively through learning and
mastering pop music. Pop music teaching in colleges and universities mainly presents
the following characteristics.
1. Both artistic and popular
The teaching of popular music in higher education should be fully based on art
education, but on top of that, it should also ensure that it is more significantly
popular to ensure that students can understand and accept it in the teaching
process. In the teaching of pop music in colleges and universities, it is because of
the artistic and popular characteristics that students are more likely to understand
and sing it. In conclusion, the teaching of pop music needs to be both popular and
artistic. In the process of development, universities should firmly grasp the pulse of
the development of art education, distinguish it from other types of music, and make
the teaching height continuously improve through more professional control.
2. Tradition and fashion intermingle
The content of popular music teaching in colleges and universities have both
characteristics of tradition and fashion. Tradition refers to the unique ideological
nature of popular music over the years of development, which can, to a certain
extent, pass on the culture and values of the era in which it lives, and has a
stronger ideological appeal and influence, and that influence is fundamental to the
continuous development of popular music, as well as being a characteristic that
needs to be observed in the process of building popular music.
3. Integrating and diversifying at the same time
The reason for this is that the curriculum should take into account the
comprehensive quality and development requirements of students, and needs to
have a certain degree of comprehensiveness, but also needs to add corresponding
special practical courses to fully ensure that students can have a high level of
comprehensive quality. In addition, pop music was established late, and it needs to
integrate instrumental music, vocal music and music management in the actual
teaching process to improve its perfection and become one of the more mature
music majors, so it has a more significant comprehensive.
https://doi.org/10.17993/3ctic.2023.121.200-220
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.42 | Iss.12 | N.1 January - March 2023
214
4. Unity and independence are common
Popular music in colleges and universities can also fully reflect the unity and
independence of music in the teaching process. Unity mainly refers to the teaching
process, teachers set the subject construction plan, curriculum system design and
other content based on the curriculum design and other music majors' teaching
methods. Based on this, the construction of each music major can maintain a high
degree of unity, and the basic construction steps in the implementation process will
not have a high degree of deviation, and can jointly promote the orderly
development of the overall music major construction.
5. Inheritance and innovation co-exist
Heritage and innovation is one of the characteristics of teaching popular music in
colleges and universities, and also the purpose of teaching popular music in
colleges and universities. Artistic, popular, traditional and fashionable belong to the
characteristics of pop music itself, while inheritance and innovation are based on
the artistic level for enhancement. The inheritance of teaching and professional
construction of popular music in colleges and universities is mainly reflected in a
variety of musical concepts, music construction and inheritance of traditional art
music, in addition to the inheritance and expansion of the way of professional
construction of conventional music and the development of the experience of art
education in colleges and universities. Compared to other contents, the way of
building conventional music is the inheritance of methods, while the way of
developing popular music mainly targets the innovation level and is the most
innovative compared to other characteristics. In actual teaching, the combination of
inheritance and innovation can fully reflect the developmental qualities of the
professional construction of popular music.
4.2. THE ROLE OF COGNITIVE-EMOTIONAL INTERACTIVE
ROBOTS IN MUSIC TEACHING
The form of education is an inherent requirement of educational work, and it is also
an inevitable requirement of society and the times for educational work. Educators in
general should make more innovative and diligent research in teaching methods and
approaches, work with diversified, multi-level and multi-angle music teaching
methods, and incorporate the healthy parts and excellent parts of the many traditional
music forms in China into the teaching contents, meanwhile, they should widely
absorb the advanced teaching experiences at home and abroad, and show the many
excellent music forms to students in the teaching process, so that they can It is also
beneficial to the promotion and growth of these excellent forms of music to have more
choices while relying on popular music.
Cognitive-emotional interactive robots can be used to create an intimate teaching
atmosphere of mutual respect, mutual trust and mutual help, thus making the teaching
relationship more pleasant and harmonious. In such a relaxed and pleasant
atmosphere, students' self-confidence will be improved and they will develop lively,
cheerful and positive psychological qualities, which can not only relieve some of the
https://doi.org/10.17993/3ctic.2023.121.200-220
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.42 | Iss.12 | N.1 January - March 2023
215
more common bad emotions such as nervousness and anxiety in the college pop
music classroom, but also increase the fun of learning pop music and turn pop music
learning into a happy and joyful thing, and face the future with a more positive They
will be able to face the present and future with a more positive attitude.
The high school pop music classroom is required to teach specific one-on-one
lessons. The teaching process begins with listening to the student's singing of the
song, then finely identifying any vocal problems that arise and giving the student
specific instructions on how to perform the song. The student then imitates and sings
according to the Emotion Robot's analysis and suggestions, while at the same time,
the Emotion Robot accurately listens to see if the sound the student is making is up to
snuff, similar or consistent with what the Emotion Robot expects, and guides the
student to find the correct vocal method for singing.
The classification results of the cognitive-emotional interaction robot are shown in
Figure 4 by recording and analyzing 15 pop music singing performances of multiple
students.
Figure 4. Vocal and musical classification
Figure 4 gives a detailed comparison of the intermediate results of classification
based on the SVM classifier and the final classification results after low-pass filtering,
as well as the percentage of human voices in each song. It can be seen that the SVM-
based classification method and the post-processing filtering mechanism are effective.
The first two songs are rap-style songs, which show strong linguistic properties, so
MFCC combined with the classical model of traditional speech recognition like SVM
has got good results on them. The classification effect of the 4th song and the 12th
song is poor, the former as a Cantonese song has a great difference in vocal
characteristics from the training data; the latter female voice is more difficult to
distinguish from the music and cannot effectively capture the difference between the
https://doi.org/10.17993/3ctic.2023.121.200-220
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.42 | Iss.12 | N.1 January - March 2023
216
human voice and music. In contrast, the classification method of the cognitive-
emotional interactive robot is effective in terms of the overall classification effect.
The cognitive-emotional interaction robot was also able to identify the emotion of
the student's pop music singing to compare it with the emotional tone of the original
song, and the results are shown in Figure 5.
Figure 5. Emotion recognition in music singing
It is not difficult to find that data points of various types of emotions are intertwined
on the vector space due to the existence of errors in the data, and since the projection
process is a linear transformation, the fusion of data with each other has a small
impact on the accuracy of the robot cognitive-emotional discrimination based on
reinforcement learning. Based on the basic rhythmic information, the cognitive-
emotional interactive robot can perform the initial recognition of basic emotions.
Although the recognition results need to be further improved, this conclusion can be
applied to musical emotion recognition systems where recognition accuracy
requirements are not very stringent. Further, we can use a neural network approach to
perform more accurate emotion information recognition using a more adequate set of
feature parameters and apply it to improve the recognition rate and robustness of the
music recognition system and achieve more human and intelligent human-robot
emotion interaction.
5. CONCLUSION
The current intelligent development of human-robot interaction systems has
reached a high level, and robotic cognitive-emotional computing has received more
and more attention from researchers as an important research component of its
intelligent development. In this paper, we carry out research work on the problems of
robot cognitive-emotional research in open-domain and closed-domain systems,
propose a robot cognitive-emotional interaction model based on reinforcement
learning, and apply the model to the innovation of university popular music teaching in
traditional music culture. The integration of Chinese traditional music culture in pop
music professional education can well cultivate students' traditional music culture
https://doi.org/10.17993/3ctic.2023.121.200-220
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.42 | Iss.12 | N.1 January - March 2023
217
literacy and traditional music culture communication ability. The combination of
modern information technology and artificial intelligence technology is in line with the
development trend of pop music, and the efficient combination of the two can inject
more contemporary vitality into pop music, stimulate the musical creativity of college
students, and improve the students' artistic professionalism and social
competitiveness.
DATA AVAILABILITY
The data used to support the findings of this study are available from the
corresponding author upon request.
CONFLICTS OF INTEREST
The author declares that there is no conflict of interest regarding the publication of
this paper.
REFERENCES
(1) Wang, Y. (2014). Fuzzy Causal Patterns of Humor and Jokes for Cognitive
and Affective Computing. International Journal of Cognitive Informatics and
Natural Intelligence, 8(2), 33-45.
(2) Yang, H., Codding, D., Mouza, C., et al. (2021). Broadening Participation in
Computing: Promoting Affective and Cognitive Learning in Informal
Spaces. TechTrends, 65(3).
(3) Hsu, D. F. (2013). Cognitive diversity in perceptive informatics and affective
computing. In IEEE International Conference on Cognitive Informatics &
Cognitive Computing (pp. 294-297). IEEE.
(4) Mercadillo, R. E., Sarael, A., & Barrios, F. A. (2018). Effects of Primatological
Training on Anthropomorphic Valuations of Emotions. IBRO Reports, 5,
54-59.
(5) Takanishi, A., Endo, N., & Petersen, K. (2012). Towards Natural Emotional
Expression and Interaction: Development of Anthropomorphic Emotion
Expression and Interaction Robots. International journal of synthetic
emotions, 3(2), 47-62.
(6) Smith, E. K. (2016). A descriptive analysis of high school choral teachers'
inclusion of popular music in current teaching practices.
(7) So, J. (2013). A further extension of the Extended Parallel Process Model
(E-EPPM): implications of cognitive appraisal theory of emotion and
dispositional coping style. Health Communication, 28(1), 72-83.
(8) Luo, X., & Min-Jiang, A. I. (2019). College Students' Cognitive Appraisal,
Emotional Identification and Reaction Mode of Online News. Journal of
Jimei University(Education Science Edition).
https://doi.org/10.17993/3ctic.2023.121.200-220
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.42 | Iss.12 | N.1 January - March 2023
218
(9) Tao, W., & Huang, Y. (2013). Research on Disposal Station Location Problem
Based on Genetic and Simulated Annealing Algorithm. In 2013 International
Conference on Computational and Information Sciences.
(10) Meng, X. (2021). Optimization of Cultural and Creative Product Design
Based on Simulated Annealing Algorithm. Complexity, 2021.
(11) Zhang, A., Wu, S., Zhang, X., et al. (2020). EmoEM: Emotional Expression in
a Multi-turn Dialogue Model. In 2020 IEEE 32nd International Conference on
Tools with Artificial Intelligence (ICTAI). IEEE.
(12) Yang, J., & Wu, C. (2021). Emotional Response Generation in Multi-Turn
Dialogue. Journal of Physics: Conference Series, 1827(1), 012124.
(13) Sun, X., Peng, X., & Ding, S. (2017).
Emotional human-machine conversation
generation based on long short-term memory. Cognitive Computation.
(14) Christ, N. M., Elhai, J. D., Forbes, C. N., Ford, J. D., & Adams, T. G. (2021). A
machine learning approach to modeling PTSD and difficulties in emotion
regulation. Psychiatry Research, 301, 113947.
(15) Kelly, S. (2009). Teaching Music in American Society: A Social and Cultural
Understanding of Music Education - Steven N. Kelly. Routledge.
(16) Schmid, E. V. (2015). Popular music in music education in Germany -
historical, current and cross-cultural perspectives.
(17) Qamash, M., Altal, S. M., & Jawaldeh, F. E. (2011). Dimensional Common
Emotional Intelligence for the Student of Higher Education In Princess Alia
College At the University of Al Balq'a Applied University In Jordan from the
Point of View of the Students. European Journal of Social Sciences.
(18) Cowie, R., Doherty, C., & McMahon, E. (2009). Using dimensional
descriptions to express the emotional content of music. In Affective
Computing and Intelligent Interaction and Workshops, 2009. ACII 2009. 3rd
International Conference on (pp. 1-8). IEEE Xplore.
(19) Selvaraj, J., Murugappan, M., Wan, K., & Yaacob, S. (2013). Classification of
emotional states from electrocardiogram signals: a non-linear approach
based on hurst. BioMedical Engineering OnLine, 12(1), 44.
(20) Martin, Schels, M., Kächele, M., Glodek, M., & Kopp, S. (2013). Using
unlabeled data to improve classification of emotional states in human
computer interaction. Journal on Multimodal User Interfaces, 8(1), 169-176.
(21) Schels, M., Kächele, M., Glodek, M., & Kopp, S. (2014). Using unlabeled data
to improve classification of emotional states in human computer
interaction. Journal on Multimodal User Interfaces, 8(1), 169-176.
(22) Weiguo, W. U., & Hongman, L. I. (2019). Artificial emotion modeling in PAD
emotional space and human-robot interactive experiment. Journal of Harbin
Institute of Technology.
(23) Zafar, Z., Ashok, A., & Berns, K. (2021). Personality Traits Assessment using
P.A.D. Emotional Space in Human-robot Interaction. In 5th International
Conference on Human Computer Interaction Theory and Applications.
(24) Song, J., Zhang, X. Y., Sun, Y., & Zhang, B. Y. (2016). Emotional speech
recognition based on PAD emotion model. Microelectronics & Computer.
https://doi.org/10.17993/3ctic.2023.121.200-220
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.42 | Iss.12 | N.1 January - March 2023
219
(25) Hsieh Y Z, Lin S S, Luo Y C, et al. (2020). ARCS-Assisted Teaching Robots
Based on Anticipatory Computing and Emotional Big Data for Improving
Sustainable Learning Efficiency and Motivation. Sustainability, 12. https://
doi.org/10.3390/su12145605
(26) Puviani L, Rama S, & Vitetta G M. (2018). A Mathematical Description of
Emotional Processes and Its Potential Applications to Affective
Computing. IEEE Transactions on Affective Computing, 9(1), 1-1. https://
doi.org/10.1109/TAFFC.2018.2887385
(27) Good J, Rimmer J, Harris E, et al. (2013). Self-Reporting Emotional
Experiences in Computing Lab Sessions: An Emotional Regulation
Perspective. PPIG. https://www.ppig.org/papers/25th-good.pdf
(28) Nelson A B, Serena R, Elisa T, et al. (2020). Neural fatigue due to intensive
learning is reversed by a nap but not by quiet waking. SLEEP, 43(4), 1-12.
https://doi.org/10.1093/sleep/zsaa143
(29) Cheng J, Sollee J, Hsieh C, et al. (2022). Correction to: COVID-19 mortality
prediction in the intensive care unit with deep learning based on
longitudinal chest X-rays and clinical data. European Radiology, 32(1), 1-1.
https://doi.org/10.1007/s00330-022-08680-z
(30) Aisling, McMahon, Gabor, et al. (2017). Intensive care microbiology pearls:
learning by 1-500-5-1. Medical Education, 51(5), 541-542. https://doi.org/
10.1111/medu.13289
(31) Nousiainen, Markku, Garbedian, et al. (2015). Toronto Orthopedic boot camp
III: Examining the efficacy of student-regulated learning during an
intensive, laboratory-based surgical skills course (vol 154, pg 29, 2013).
Surgery, 158(6), 1756-1757. https://doi.org/10.1016/j.surg.2013.05.003
(32) Panksepp J, & Watt D. (2011). What is Basic about Basic Emotions? Lasting
Lessons from Affective Neuroscience. Emotion Review, 3(4), 387-396.
https://doi.org/10.1177/1754073911410741
(33) Gilead M, Katzir M, Eyal T, et al. (2016). Neural correlates of processing "self-
conscious" vs. "basic" emotions. Neuropsychologia, 80, 207-218. https://
doi.org/10.1016/j.neuropsychologia.2015.12.009
(34) L., Wenling (2023). Deep Learning Network-Based Evaluation method of
Online teaching quality of International Chinese Education. 3C Tecnología.
Glosas de innovación aplicada a la pyme, 12(1), 87-106. https://doi.org/
10.17993/3ctecno.2023.v12n1e43.87-106
(35) Liu Hailiang, Hou Chenglong, & Ramzani Sara Ravan. (2021). Construction
and reform of art design teaching mode under the background of the
integration of non-linear equations and the internet. Applied Mathematics
and Nonlinear Sciences, 7(1), 215-222. https://doi.org/10.2478/
amns.2021.2.00149
https://doi.org/10.17993/3ctic.2023.121.200-220
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.42 | Iss.12 | N.1 January - March 2023
220